A new class of alternating proximal minimization algorithms with costs-to-move

被引:59
作者
Attouch, H.
Redont, P.
Soubeyran, A.
机构
[1] Univ Montpellier 2, Inst Math & Modelisat Montpellier, CNRS, UMR 5149, F-34095 Montpellier, France
[2] Univ Aix Marseille 2, CNRS, GREQAM, UMR 6579, F-13290 Les Milles, France
关键词
alternating minimization; alternating projection; proximal algorithms; costs to change; inertia; anchoring effect; dynamical game theory; Nash equilibria; steepest descent; dissipative dynamical systems; MONOTONE-OPERATORS; DYNAMICAL-SYSTEM; VARIATIONAL-INEQUALITIES; POINT ALGORITHM; HILBERT-SPACE; CONVERGENCE; SUM; PROJECTION; FRICTION; ZERO;
D O I
10.1137/060657248
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
Given two objective functions f : X -> R boolean OR{+infinity} and g : Y -> R boolean OR{+infinity} on abstract spaces X and Y, and a coupling function c : X x Y -> R+, we introduce and study alternative minimization algorithms of the following type: ( x(0), y(0)) is an element of X x Y given; (x(n), y(n)) -> ( x(n+1), y(n)) -> (x(n+1), y(n+1)) as follows: x(n+1) is an element of argmin{f(xi) + beta(n)c(xi, y(n)) + alpha(n)h(x(n), xi) : xi is an element of X}, y(n+1) is an element of argmin{g(eta) + mu(n)c(x(n+1), eta) + nu(n)k( y(n), eta) : eta is an element of Y}. Their most original feature is the introduction of the terms h : X x X -> R+ and k : Y x Y -> R+ which are costs to change or to move (distance- like functions, relative entropies) accounting for various inertial, friction, or anchoring e. ects. These algorithms are studied in a general abstract framework. The introduction of the costs to change h and k leads to proximal minimizations with corresponding dissipative effects. As a result, the algorithms enjoy nice convergent properties. Coefficients alpha(n), beta(n), mu(n), nu(n) are nonnegative parameters. When taking alpha(n) = nu(n) = 0 and quadratic costs on a Hilbert space, one recovers the classical alternating minimization algorithm, which itself is a natural extension of the alternating projection algorithm of von Neumann. A number of new signi. cant results hold in general metric spaces. We pay particular attention to the following cases: (1.) (X, d(X)) and (Y, d(Y)) are complete metric spaces and h >= d(X), k >= d(Y) ("high local costs to move"); the algorithms then provide sequences that converge to Nash equilibria. (2.) X = Y = H is a Hilbert space, the costs to change are quadratic ("low local costs to move") and the functions f, g : H -> R boolean OR{+infinity} are closed, convex, proper; then some of the classical convergence theorems for alternating convex minimization algorithms, including those of Acker and Prestel, are properly extended with original proofs.
引用
收藏
页码:1061 / 1081
页数:21
相关论文
共 39 条
[1]  
Adly S, 2006, ADV MECH MATH, V12, P289
[3]   An inertial proximal method for maximal monotone operators via discretization of a nonlinear oscillator with damping [J].
Alvarez, F ;
Attouch, H .
SET-VALUED ANALYSIS, 2001, 9 (1-2) :3-11
[4]  
[Anonymous], 1997, Contemporary Mathematics
[5]  
[Anonymous], 1980, Ann. Fac. Sci. Toulouse V. Ser. Math.
[6]  
[Anonymous], ACTA SCI MATH SZEGED
[7]  
[Anonymous], 1950, ANN MATH STUDIES
[8]  
Attouch A, 2005, MOS-SIAM SER OPTIMIZ, V6, P1
[9]   Regularized Lotka-Volterra dynamical system as continuous proximal-like method in optimization [J].
Attouch, H ;
Teboulle, M .
JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS, 2004, 121 (03) :541-570
[10]   The heavy ball with friction method, I. The continuous dynamical system: Global exploration of the local minima of a real-valued function by asymptotic analysis of a dissipative dynamical system [J].
Attouch, H ;
Goudou, X ;
Redont, P .
COMMUNICATIONS IN CONTEMPORARY MATHEMATICS, 2000, 2 (01) :1-34